▶Terence Tao consistently views AI as a powerful tool for the 'experimental' or 'idea generation' phase of science, drastically lowering the cost of producing hypotheses and enabling large-scale data gathering on technique effectiveness (Claims 3, 4, 17).Apr 2026

▶Tao believes the future of mathematics lies in hybrid human-AI teams, where human depth complements AI's breadth, asserting that a full replacement of human mathematicians is not imminent and would require new breakthroughs (Claims 5, 6).Apr 2026

▶Tao maintains that current AI models have significant limitations for core mathematical work, including a very low success rate on novel problems (1-2%), an inability to build on progress cumulatively, and a failure to speed up the most difficult parts of problem-solving (Claims 1, 10, 19).Apr 2026

▶There is a tension between Tao's assertion that he still uses pen and paper for the hardest problems and Nathan Labenz's report that an AI solved a challenging problem posed by Tao in weeks, a task that took humans 18 months to make progress on (Claims 1, 8).Apr 2026

▶Tao highlights both a recent surge and a subsequent pause in AI's ability to solve Erdős problems, noting that after solving about 50, progress has stalled as the 'low-hanging fruit' has been picked, suggesting a potential performance plateau (Claims 12, 21).

▶Tao predicts AI will automate the bulk of mathematicians' work within a decade, yet simultaneously states that current AI tools have only a 1-2% success rate in systematic studies, indicating a significant performance gap that must be closed to realize his prediction (Claims 2, 19).Apr 2026

▶While Tao celebrates that AI has driven the cost of idea generation to near zero, he also warns this has created a new bottleneck in verification and is flooding scientific journals with low-quality, AI-generated papers (Claims 4, 17, 20).Apr 2026